DI

SC

US

SI

ON

P

AP

ER

S

ER

IE

S

Forschungsinstitut zur Zukunft der ArbeitInstitute for the Study of Labor

Global Reform of Personal Income Taxation,1981-2005: Evidence from 189 Countries

IZA DP No. 4228

June 2009

Klara Sabirianova PeterSteve ButtrickDenvil Duncan

Global Reform of Personal

Income Taxation, 1981-2005: Evidence from 189 Countries

Klara Sabirianova Peter Georgia State University,

CEPR and IZA

Steve Buttrick Georgia State University

Denvil Duncan

Georgia State University

Discussion Paper No. 4228 June 2009

IZA

P.O. Box 7240 53072 Bonn

Germany

Phone: +49-228-3894-0 Fax: +49-228-3894-180

E-mail: iza@iza.org

Any opinions expressed here are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but the institute itself takes no institutional policy positions. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit organization supported by Deutsche Post Foundation. The center is associated with the University of Bonn and offers a stimulating research environment through its international network, workshops and conferences, data service, project support, research visits and doctoral program. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.

IZA Discussion Paper No. 4228 June 2009

ABSTRACT

Global Reform of Personal Income Taxation, 1981-2005: Evidence from 189 Countries

In this paper we use a panel of 189 countries to describe the salient trends that have emerged in national personal income tax systems spanning the twenty five year period from 1981 to 2005. Using complete national income tax schedules, we calculate actual average and marginal tax rates at different income levels as well as time-varying measures of structural progressivity and complexity of national tax systems. We show that frequent alterations of tax structures have reduced tax rates at higher levels of income and diminished the overall progressivity and complexity of national tax systems; however, the degree of this change varies considerably across countries. We also find that the relationship between the tax rates and revenue is positive for high income countries; however, the strength of the relationship declines with weaker institutions and lower levels of economic development. JEL Classification: C8, E62, H2, H87, N10, O1 Keywords: personal income tax, marginal rate, average rate, tax complexity,

progressivity, flat tax, revenue, global trends Corresponding author: Klara Sabirianova Peter Department of Economics Andrew Young School of Policy Studies Georgia State University P.O. Box 3992 Atlanta, GA 30302-3992 USA E-mail: kpeter@gsu.edu

2

1. Introduction

About one hundred years ago, Seligman (1908) observed that “the tendency toward

progressive taxation is almost everywhere on the increase”. However, income tax progressivity

at the beginning of the 20th century was rather moderate by contemporary standards. The top

personal income tax (PIT) rate in 15 surveyed countries hardly exceeded 10%. “The highest

point known to history as actually enforced is thirty-seven and a half percent”, which lasted only

a year in Holland in 1796. Referring to this case, Seligman remarks “the progression was so

severe as to become a confiscation”.

By 1981, the 37.5% top PIT rate was no longer the highest historical point – 4 of every 5

country had the top PIT rate exceeding 37.5%, with the maximum of 90% in Iran. The GDP-

weighted average top statutory marginal PIT rate was 62% among 108 countries. National PIT

systems were frequently plagued by multiple tax schedules, complicated tax formulas,

surcharges, numerous exemptions, and escalating highly-partitioned tax scales.

However, many governments made significant changes to their national PIT systems,

since 1981. The GDP-weighted average top statutory marginal PIT rate fell from 62% in 1981 to

42.9% in 1991 and by 2005 reached a twenty five year low of 36.4%. There was also a sizeable

drop in marginal and average PIT rates at higher levels of individual income and an overall

decline in structural progressivity of the PIT systems. We observe a growing trend toward the

use of one-rate flat personal income taxation1 and find some evidence of the increased simplicity

of several characteristics of the PIT structure. It is apparent that the current tax rates are lower

while tax structures are flatter and, to a certain degree, simpler than they were 25 years ago.

1 Although this trend is more popular among transitional countries at the moment, there is ongoing debate in other parts of the world about the merits of making the switch (Paulus and Peichl 2008).

3

This study adds to the taxation literature by providing a careful methodological

documentation of the changes that have taken place in personal income taxation over the last 25

years. The analysis we provide is important for a number of reasons. First, the coverage of the

analysis is unprecedented as we describe the changes that have taken place in almost 190

countries worldwide. Relying on such a large panel of countries means that the trends we

describe are not likely to suffer from a sample selection bias. More importantly, we are able to

document the variation in the personal income tax (PIT) evolution across countries at different

stages of economic development.

Second, we calculate and analyze the time varying marginal and average tax rates at

different points of the income scale. These tax measures represent a significant improvement

over the commonly used top statutory PIT rate. The existence of deductions, allowances, tax

credits, surtaxes, and local taxes means that there is often a big difference between the effective

and statutory tax rates. This, along with excessive income inequality, implies that the top rate

does not apply to most individuals in many countries. We address this issue by using more

common levels of earned income in calculating tax rates and by making adjustments for

deductions, allowances, tax credits, and other legal rules. As such, our tax rate measures are

much closer to the actual rates that individuals are supposed to pay.

Third, the unique information we collected allows us to identify the long-term trends in

such important features of national PIT systems as progressivity and complexity. Our

progressivity measure is the progression slope of either average or marginal rates along the

income scale, which is calculated for each country and year using 100 data points that are formed

around the country’s GDP per capita. This measure has been trending downward in every

country category, with the exception of low income countries. We also summarize the trends in

4

several observable dimensions of the PIT complexity, including the complexity of allowances

and tax credits, the use of multiple tax schedules, non-standard tax formulas, local taxes, national

surtaxes, and the number of PIT brackets.

Fourth, the existing measures of PIT rates are generally available for a cross section of

countries at a point in time or very short panels with limited time variation. Furthermore, the

existing data sources usually under represent developing countries. The coverage and

comparability of our tax measures go a long way in filling this gap. Because we rely on data

from several sources, we are able to provide a greater level of consistency, both across countries

as well as within country over time, than data from any one source. Having a long panel is also

useful in identifying within-country dynamics of tax measures while controlling for constant

cross-country heterogeneity. We use a simple growth model to illustrate how having a large

representative panel of countries with consistent time-series may have a significant impact on the

inference and policy conclusions regarding the effect of tax rates on economic growth.

Finally, we examine the relationship between the tax rates and collected revenue for a

large subset of countries. The question we are interested in is whether the reduced rates and

progressivity had a considerable negative effect on PIT collection. We find that the relationship

between the tax rates and revenue is positive for high income countries; however, the strength of

the relationship declines with weaker institutions and lower levels of economic development.

The PIT rates do not seem to matter for revenue collection in countries with high levels of

government corruption.

Admittedly, we ignore other components of the tax system such as taxes on corporate

income, domestic consumption, and international trade. Accounting for changes in these other

taxes is important when looking at the impact of the global reform on economic outcomes.

5

However, collecting and analyzing information on these additional components, while useful,

would prove too monumental a task for one paper. We leave the analysis of other components to

future research.

We begin by providing a quick overview of the data followed by an analysis of major

global trends in personal income taxation, in particular, the high frequency of PIT changes, a

ubiquitous decline in top statutory PIT rates, a downward movement in actual marginal and

average rates at higher levels of individual income, increasing simplicity of national tax systems,

decreasing structural PIT progressivity, and the emergence of flat rate tax schedules. This is

followed by an analysis of the effect these changes may have had on PIT collection. We then

discuss the sensitivity of estimation results to the size and composition of the sample of

countries.

2. Data Sources and Sample

The data analyzed in the paper covers personal income tax structures in 189 current and

former countries worldwide for the period 1981 to 2005.2 As far as we know, this is the first

panel study of this size that analyzes the historical changes that have occurred in personal

income tax structures over the last 25 years.

Our dataset is comprehensive and contains the complete national PIT schedules with

statutory rates, tax brackets, country-specific tax formulas, basic allowances, standard

deductions, tax credits, multiple tax scales, national surcharges, and local taxes. This

information allows us to compute several important variables including country’s statutory rates,

number of tax brackets, actual average and marginal tax rates at different points of income

distribution, frequency of tax changes, and various measures of PIT progressivity and 2 The number of countries ranges from 108 in 1981 to 180 in 2004.

6

complexity. In the following sections we give a detailed description of each variable, including

definitions, computation methodology, summary statistics, and trend analysis.

This consolidated PIT dataset was assembled from more than 100 distinct reference

books and datasets. Of these, the most important sources are the tax anthologies published by

international accounting firms (e.g., PricewaterhouseCoopers, Coopers and Lybrand, etc.) as well

as publications and datasets of international organizations (e.g., International Bureau of Fiscal

Documentation, OECD, World Bank, International Monetary Fund, etc.) and public policy

centers (such as University of Michigan’s World Tax Database (WTD), Economist Intelligence

Unit, Heritage Foundation, and Fraser Institute). A wide variety of data sources allows for

significant cross checking, thus minimizing the amount of errors that one will inevitably

encounter in a study of this magnitude. We made every effort to reconcile contradictory

reporting to produce a consistent series for each country across time (see Appendix 1 for the

description of sources and data reconciliation efforts).3 Through careful cleaning of the data, we

have assembled a valuable consolidated personal income tax resource for scholars doing cross-

country empirical research.4

One notable advantage of this consolidated dataset is that it provides a level of world

representation that has not been attained in previous studies of tax structures.5 Our full sample

includes 3613 top statutory PIT rates for 189 current and former countries from 1981 to 2005.

3 The following examples can give a sense of inconsistencies we dealt with. The Russian top PIT rate is coded as 90% in the 1990 WTD while this rate was applied only to the inheritors of book writers for honorariums received. We had to change this rate to the second highest rate of 60% charged on incomes from individual economic activity that was often performed underground. The top marginal PIT rate for Denmark rises from 22% to 68% in the 1988 WTD while in fact it drops from 73% to 68%. The reason for such a discrepancy is omitted surcharges and regional taxes prior to 1988 in WTD. Significant inconsistencies may affect the overall trends as well as bias the estimates in the behavioral models based on these data. 4 This dataset is planned to be publicly released as part of the World Tax Indicators v.1 in 2009. 5 Some of the largest samples in earlier studies include 51 marginal PIT rates from developing countries in 1984-1985 (Sicat and Vermani 1988) and 66 top PIT rates in 1980-1989 (Lee and Gordon 2005).

7

This is an average of 145 countries per year. Even in the case of highly data-demanding

variables such as tax progressivity measures which require the complete tax schedule, we have

non-missing values for 175 countries, or an average of 123 countries per year (from 75 in 1981

to 157 in 2002). To further highlight the representativeness of our sample, we note that 31% of

the countries included are classified as high income countries, 18% as upper middle income,

30% as lower middle income, and 22% as low income countries.6 These countries are located in

all populated geographic regions and continents,7 and they represent approximately 94% of the

world population and 98.5% of the world output (or more precisely, 86% of the world GDP in

current U.S. dollars in 1981, 89% in 1982, 97% in 1983-1988, and 98-99% in all subsequent

years).8

3. Continuous Tax Reform

An examination of the dataset reveals a number of remarkable changes in personal

income taxation over the last twenty five years. The first important observation is that tax rates,

income thresholds, and the overall tax structures are continuously changing. The last column of

Panel A in Table 1 shows that, on average, 45% of the countries in our sample change at least

one element of their national tax scale (statutory rates or tax brackets) every year. Of these

countries, about half make full changes by altering both rates and bracket thresholds

simultaneously. The share of countries with full changes in PIT was particularly high in the late

1980s relative to other periods. We find very few cases where countries modify their statutory

6 Country categories are defined using the World Bank country classification based on historical (time-varying) income thresholds. For example, the income thresholds used for the 2005 classification are as follows: low income, $875 or less, upper middle income, $876-$3465, upper middle income, $3466-$10725, and high income, $10725 or more. 7 The regional breakdown is as follows: Africa – 48 countries, Asia and Oceania – 31, Central and South America – 35, Eurasia – 16, Europe – 43, Middle East – 12, and North America – 4. 8 The major economies with missing PIT rates in the early 1980s are former USSR in 1981-1982 and Italy in 1981.

8

rates while maintaining old thresholds (2.7% of the sample). However, the share of countries

that adjust tax brackets but keep their statutory rates the same increased significantly in the early

1990s moving from 15% in the 1986-1990 period to 21% in the 1991-1995 period. Changes of

this nature might be an indication that countries are increasingly adjusting their tax schedules for

inflation.

An investigation of these changes by country type reveals that national PIT scales are

changing more frequently in developed countries (62% of countries in the respective category

per year) followed by upper middle income countries (44%), lower middle income countries

(42%), and low income countries (29%). The significant number of developed countries that

went through major tax reforms in the late 1980s explains the spike in the share of high income

countries with full changes in their tax scales during the 1986-1990 period (42.7%).

The difference between high and low income countries is especially striking with respect

to the share of countries that alter thresholds but not rates. From Table 1 it becomes apparent

that most of the low income countries do not adjust their tax brackets for inflation over the whole

considered period.9 This is also true for the middle income countries in the 1980s. Without such

inflationary adjustments, the tax structure may become less equitable if taxpayers are being

pushed into higher tax brackets with no accompanying increase in real income. This is known as

the “bracket creep” effect. Even if top statutory PIT rates are falling, the core taxpayers may

actually experience an increase in marginal rates in an inflationary environment that is

commonly present in less developed economies.

Apart from the inflationary adjustments, another explanation for continuous changes in

tax rates and brackets could be that countries are simply attempting to identify the best tax 9 The slow adjustment of tax brackets to inflation in low income countries is likely to be related to widespread evasion, weak enforcement, and little revenue collected from personal income tax.

9

structure. Identifying the best tax structure, however, is a normative matter, which can affect the

changes in different ways. Countries that are concerned about equity, for example, may want to

increase their tax rates while those more concerned about efficiency may reduce them. Whatever

the motivation behind the high frequency changes, however, one might expect some negative

feedback for efficiency. This follows from the fact that a constantly changing tax schedule adds

uncertainty to the tax system which is likely to increase complexity and reduce efficiency.10

The important question for us is whether these changes are simply transitory in nature or

is there some well defined direction of change. If changes are being made on the grounds of

efficiency, then one may expect persistent trends to develop over time as countries try to improve

the efficiency of their tax structures by reducing rates. The remainder of the paper will give a

detailed analysis of the resulting trends that have developed over time.

4. Top Statutory PIT Rates

One of the frequently cited variables of any income tax system is the top statutory

personal income tax (PIT) rate. This is a legally determined marginal tax rate applicable to the

top bracket of the income tax schedule. This particular tax rate has occasionally been used in

empirical cross-country research as a proxy variable for tax progressivity or as a way of

assessing the overall excess tax burden (e.g., Johnson, Kaufmann, and Zoido-Lobaton 1998,

Friedman et al. 2000, among others).

Figure 1 illustrates how the average top statutory tax rate has dropped considerably over

the last twenty five years. This decline was especially pronounced for the GDP-weighted trend

10 Constantly changing features of the tax schedule cause individuals to be uncertain about their tax liability and thus increase compliance and administrative costs (Slemrod 1992).

10

which is more representative of the world economy.11 One interesting observation is that the

most dramatic change occurred during the late eighties and early nineties. The weighted average

top statutory PIT rate fell from a high of 62% in 1981 to 56% in 1986. During the ensuing eight

year period (1986 - 1993), however, the PIT rate plunged by approximately 16 percentage points.

The PIT rate then increased by a modest 2 percentage points before resuming its downward trend

in 1996. Since then, the decline has continued, with average top statutory rates sliding a further

6.5 percentage points over the next 10 years. Together these changes represent a sizeable 41.2%

decline in the weighted top PIT rate from a high of 62% in 1981 to a low of 36.4% in 2005.

Further evidence in support of the worldwide downward trend is reported in Table 2

where we observe that only 17% of unweighted top PIT rates were in excess of 40% in 2001-

2005 compared to over 71% during the early 1980s. The share of countries with top PIT rates in

excess of 60% plummeted from about a quarter in 1981 to less than 1% in 2005. The lower non-

zero rates (1% to 40%), on the other hand, became more popular as the percentage of countries

falling into these categories increased from approximately 15% to over 73% between 1981 and

2005.

The reader will notice that there are two changes in the weighted PIT trend that align

perfectly with well known tax changes in the U.S. These are the 1986 and 1993 tax reforms

which reduced and increased the top PIT rates respectively. The weighted top PIT rate fell by

approximately 6 percentage points between 1986 and 1987. Similarly, the 2 percentage point

increase in the top PIT rate between 1993 and 1994 coincides with the U.S. Budget

Reconciliation Act enacted in August 1993. This casual observation would imply, incorrectly,

that the larger countries, as measured by GDP, are driving these trends with little or no change 11 The values are weighted by gross domestic product measured in 1990 U.S. dollars. Underlying numbers are reported in Appendix Table A1.

11

taking place elsewhere in the world. However, the summary statistics presented in Table 3 and

Figure 2 show clearly that the changes are ubiquitous. From Table 3, for example, we are able to

highlight that each country category experienced a significant decline in the GDP-weighted top

PIT rate. The greatest decline was among the upper middle income countries, followed by low,

lower middle and high income countries respectively. By 2005, upper middle income countries

had the lowest top PIT rates on average. The downward trend remains statistically significant

after controlling for country fixed effects and accounting for serial correlation.

While the evidence in Panel A proves that the downward movement in top PIT rates was

universal, the trend coefficients in Panel B also suggest that the major changes appear to be

sequential. What started as top tax rate cuts in high income countries during the early and mid

1980s was spread to middle income countries during the late 1980s and early 1990s before

eventually reaching low income countries in the late 1990s.

5. Actual Marginal and Average PIT Rates

The top statutory PIT rate is generally quoted in impressionistic comparisons of national

income tax schedules, and, in particular, as a proxy for the tax burden in some empirical

research. In a sense, it is easy to see why this rate is so widely used, as it is potentially the

maximum marginal tax rate facing the wealthiest taxpayers. However, the top statutory rate in

less developed economies is often irrelevant to the majority of domestic high earners and

entrepreneurs, with the exception of the small number of expatriates and wealthiest local elite.

For example, in high income countries the top statutory PIT rate is applicable, on average, to the

level of income equivalent to the triple of a country’s GDP per capita. In contrast, the ratio of

the top threshold to a country’s GDP per capita is about 18 in upper middle income countries, 47

in lower middle income countries, and 83 in low income countries. Thus, it is important to

12

examine the trends in PIT rates at other points of the income distribution that are more relevant

to the majority of population. The problem though, is that it is practically unfeasible to find the

actual income distribution for the large number of countries over the 25-year period. As an

alternative, we use a country’s GDP per capita and its multiples as a comparable income base.

For example, for each country in our sample, we can calculate the marginal rates for the level of

individual income equivalent to 1, 2, 3, and 4 times GDP per capita by using the national PIT

scale.12

In addition to having rates at different levels of individual earnings, our data also allow us

to calculate the actual marginal tax rates that are generally more preferred to the statutory rates

dictated by national tax schedules. It is not unusual for these two rates to differ significantly. A

local tax, national surtax, or additional tax schedules can raise the actual marginal tax rate above

the officially stated rate. Personal deductions and tax credits, on the other hand, can lower the

actual marginal tax rate below the legal rate. To estimate actual marginal tax rates for each

country in our sample, we first calculate taxable income for 100 different levels of pre-tax

income that are evenly spread in the range from 4% to 400% of a country’s GDP per capita.

Taxable income excludes standard deductions, basic personal allowances, and employee/wage

allowances, which are unconditionally applicable to all single, employed taxpayers.13 Next, we

apply the tax schedule and, whenever relevant, particular tax formulas to the taxable income in

12 This approach was also used by Sicat and Vermani (1988) in comparing marginal rates across 51 developing countries. 13 We use the rates for single taxpayers in calculating taxable income to preserve comparability across countries and because most of countries have individual-based tax systems. Some countries, like the U.S., Germany, Zimbabwe, have income tax systems based on the combined family income. However, because the number of countries with separate rates for married individuals is small in our data (8 to 12 in various years), it is unlikely that using rates for married taxpayers would alter the aggregate trends.

13

order to compute the tax liability for each income level.14 The tax liability also includes local

taxes and major national surtaxes if altogether they exceed 5% of the taxable income.15 The final

tax liability figure is adjusted by subtracting the tax credits that are universally applicable to

single taxpayers. Finally, the actual marginal rates are calculated as Δ tax liability / Δ income for

each of the 100 values of gross income.16

Using tax liability figures, we also compute the average tax rate as the ratio of total tax

liability to gross income for each level of income. It is important to emphasize that the average

rate is the tax rate that the individual taxpayers face or are supposed to pay under current tax

laws, not the rate they choose to pay by means of underreporting their income. As such, the

average tax rate gives an upper bound on the effective rate paid by individuals under the PIT.

Figure 3 shows that in any given year the marginal rates are higher than their respective

average rates and that both average and marginal rates increase with the level of income. These

are the two main indications that the personal income tax structure for most of the countries

remains progressive. More important for the current discussion, however, are the observed

trends for the various rates. Irrespective of the rate chosen, one observes a clear downward trend

over the sample period. Furthermore, all of the variables show that the greatest downturn in the

14 In most cases, we apply the standard formula for calculating tax liability, T=(Y-Yr)*tr+R, where Y is taxable income, Yr is the largest bracket less than Y, R is the tax liability if taxable income were equal to Yr and tr is the marginal tax rate for the rth bracket. However, in 418 country-year cases we had to use non-standard country-specific procedures in computing tax liability. Countries may have local taxes and multiple surtaxes, an additional schedule for salaries (Egypt, Senegal, etc.), or non-standard tax formulas (e.g., Cote d’Ivoire, Denmark, and Germany). For example, federal tax liability in Denmark is calculated as T= Y*t0+(Y-Y1)*t1+(Y-Y2)*t2, where t0 is the bottom rate, t1 is the middle rate, t2 is the top rate, Yi is the threshold for each rate and the combined top marginal rate is limited to 59% as of 2008. 15 The methods used to incorporate local taxes vary from using simple average rates (e.g., Denmark, Finland, and Sweden) to applying separate local tax schedules such as in Canada (Ontario), Japan, and Switzerland (Zurich). 16 For the level of gross income equivalent to a country’s GDP per capita, actual and statutory marginal rates are approximately the same in 66% cases (within one percentage point deviation), the statutory rate is greater in 29% cases, and it is less in 5% cases. The similarities between actual and statutory rates tend to increase with the level of gross income.

14

tax rates took place during the mid to late eighties, with a consistent but moderate decline in the

subsequent years.

Also important to note is that there is some amount of convergence between the actual

rates for higher and lower income levels. That is, while both high-income rates and low-income

rates trended downward throughout the sample period, the high-income rates (for the income

corresponding to quadruple of a country’s GDP per capita) fell at a much faster rate. For

example, marginal tax rates (MTR) at y, where y denotes GDP per capita, fell by -0.49

percentage points per year whereas MTR at 4⋅y went down by -0.76 percentage points per year;

the difference in the rates of decline is statistically significant at all conventional levels of

significance. A similar convergence result is obtained with respect to the annual decline in

average rates (-0.32 percentage point decline at y vs. -0.57 at 4⋅y). This apparent convergence

could be an indication that the tax reforms have reduced the progressivity of income tax

structures that we examine further below.

Thus, if we take a worldwide weighted average, we observe a tendency of national tax

systems toward lower actual marginal and average PIT rates at different points of the generated

income distribution, and this tendency seems particularly strong at the higher levels of individual

earnings. The unweighted trend is also negative and statistically significant even after

controlling for country time-invariant heterogeneity and serial correlation (Table 4, Panel B).

While it is true on average, this conclusion does not hold for the poorest countries. In Table 4

we report the estimated trend coefficients by country category at four income levels. High and

upper middle income countries have declining actual PIT rates over the 25-year period. In

concordance with an earlier observation, the decline in both marginal and average rates was

stronger for higher levels of individual earnings. Perhaps the most interesting result in Table 4 is

15

that the least developed countries had either no change or an increase in rates throughout the

sample period. These results are not that surprising in light of the evidence presented in the

previous section on the frequency with which the PIT structure changes. Since the low income

countries make very few changes to their bracket thresholds and experience much higher

inflation than the other groups of countries, it is predictable that the actual rates are increasing

over time, especially for the upper middle class.17 Households with below median income are

often not liable (and not able) to pay any taxes because their income is generally lower than the

first tax threshold. Therefore, entrepreneurs and high skilled workers are most likely to be

victims of the increase in average and marginal rates in the poor countries.18

Thus, the trends discussed so far make it clear that the observed downward movement in

actual PIT rates is influenced significantly by more developed economies.

6. Trend Toward Simplicity

In this section, we show that in addition to the downward trend in statutory and actual

PIT rates, there has also been a consistent drive toward PIT simplicity among the countries of the

world. Having a complex tax structure is costly as it is likely to encourage tax evasion and

unofficial economy transactions. There are also direct social costs imposed on the society by

having a complex tax system, including the costs of filing the returns (both time and monetary),

efforts to calculate deductions and allowances as well as other efforts to reduce taxable income.

Slemrod and Sorum (1984) show that these direct costs were very high for the U.S. PIT system

17 We found a positive correlation between the number of years since last change in brackets and the change in upper average and marginal tax rates (ATR and MTR at 3.y and 4.y), and that this correlation increases with the inflation rate. This finding supports the “bracket creep” explanation of increasing average and marginal rates in an inflationary environment. 18 Higher rates on entrepreneurs could be detrimental to productivity growth and discourage saving and domestic investment. Instead, high earners find ways to avoid saving and domestic investment through immigration, money hoarding, spending money on luxury imported items, and investing abroad (Hall and Rabushka 1985).

16

in 1982 consuming approximately 5−7% of tax revenues generated in that year. There is also a

tradeoff between simplicity and other characteristics of the tax structure. This is true especially

where efforts to distinguish among taxpayers in achieving a more equitable distribution increases

the complexity of the tax structure (Slemrod 1992). Therefore, the trend in complexity is

important since it has implications for both equity and efficiency.

Measuring the extent of complexity in a tax system is extremely difficult, especially in

the cross-country setting. We recognize that our cross-country analysis cannot account for many

important dimensions of tax complexity such as clarity of tax laws and regulations, costs of tax

planning and paperwork, time needed to submit declarations, number of exemptions and

itemized deductions, and other analogous factors, which are hard to measure even for one

country, much less for many countries. It is virtually impossible to collect time-varying data on

these factors for our sample; 189 countries for 25 years. Despite these limitations, our discussion

does provide useful and comparable cross-country information on several key dimensions of PIT

complexity that we are able to quantify for a large number of countries.

Among those dimensions are the complexity of allowances and tax credits, multiple tax

schedules, non-standard tax formulas, local taxes, national surtaxes, and the number of PIT

brackets. Having no allowances, for example, is considered to be less complex than having

allowances, while having allowances that depend on the spouse’s employment participation is

considered more complex than having a standard allowance.19 Similarly, having a single PIT

schedule is simpler than having multiple schedules since a single tax schedule eliminates any

19 Personal allowances and tax credits are considered standard if they are applicable to the taxpayer, spouse, and/or children without any restricting conditions (such as age, children being at school, spouse not working, etc.).

17

confusion as to which schedule individuals should use when filing their tax returns.20 Non-

standard formulas for tax computations also add to the tax complexity as they often require extra

rules, regulations and definitions to make these formulas work (footnote 16 gives an example of

non-standard tax formulas). National surtaxes and separate local/provincial PIT rates are other

obvious factors contributing to complexity. Finally, the excessive partitioning of personal

income into multiple tax brackets is often associated with larger administrative and information

costs and may create additional incentives for manipulating taxable income to move down the

graduated scale.21

Table 5 provides summary statistics for the above indicators of PIT complexity. We see

that 53% of countries had non-standard allowances or tax credits in the first five year period of

the sample, but this number decreased to 40% in the last five year period. The largest changes

took place in the first two periods where the share of non-standard allowances fell by 9% and

14% respectively. There is also a noticeable decline in the use of surtaxes and high local tax

rates (defined as 5% of taxable income or greater) and a corresponding increase in the use of low

local tax rates. Likewise, the percentage of countries using non-standard tax formulas fell by 10

percentage points or over 50% between 1981 and 2005. Unlike the other components, the use of

multiple schedules increased slightly between the first and second period before also showing a

steady decline from 22% in the second period to 17% in the last.

Results in Table 5 also indicate a significant cross-country variation in the characteristics

of the PIT systems. From the probit estimates in Panel B, we see that high income countries are

20 Multiple tax schedules include, but not limited to, alternative minimum tax as well as schedules that depend on occupation, industry, location, marital status, and other characteristics of a resident taxpayer. 21 The large number of brackets makes it difficult for taxpayers to know their marginal tax rate and hence to make more informed decisions (Slemrod 1992). It also increases the incentives of taxpayers to play around the multiple thresholds in attempt to move down the graduated scale.

18

more likely to have non-standard allowances and tax credits, high local taxes, and non-standard

tax formulas. Low income countries, on the other hand, have a higher likelihood of using

multiple tax schedules, imposing low local taxes, and using standard allowances and tax credits.

The downward trend depicted by the probit estimation is consistent with the above discussion of

the mean values by period. Panel C of Table 5 reports the estimated trend coefficient from the

linear probability model that controls for country fixed effects and accounts for arbitrary serial

correlation. While the overall trend remains negative, it becomes statistically insignificant for

three indicators (surtaxes, high local taxes, and multiple tax schedules) that could be due to

limited within-country variation in these measures.

From Table 6, we find that approximately 50% of the sample had more than 10 tax

brackets in the 1981–1985 period with approximately 9% having more than 20 brackets, up to 68

in Guatemala. This is in comparison to 5.5% and 0.4% in the 2001-2005 period, respectively.

While the 6-9 range gained marginally, increasing by a total of 3 percentage points, the 1-3 and

4-5 ranges became the norm. As of 2005, approximately 55% (23% and 32% respectively) of

countries fell into these ranges, while the zero range experienced a decline of approximately 5

percentage points. Both unweighted and GDP-weighted trends in Figure 4 indicate that the steep

decline in the average number of tax brackets was worldwide. Although there was significant

variation in the number of brackets at the beginning of the sample period, the average number of

brackets since 1995 has remained around 4-6 for all country categories.

Thus, we find that the observed indicators of PIT complexity have exhibited downward

trends over the last quarter century. The steep drop in the average number of tax brackets, the

19

gradual elimination of surtaxes, and the dwindling use of multiple tax schedules, non-standard

allowances and tax formulas has moved countries toward simpler PIT structures.22

7. Structural Progressivity

The evidence so far shows that there has been a significant shift towards more simple

income tax structures and lower tax burden, especially at the top of the income distribution.

These changes can justifiably be assumed to have a negative effect on the structural progressivity

of the income tax system. The term structural progressivity is introduced by Musgrave and Thin

(1948) to denote changes in the calculated (nominal) tax burden along the income distribution, as

opposed to the effective progressivity that depicts changes in actual income inequality.

In this study, we focus on the structural progressivity of personal income taxes. We

believe that the measures of structural progressivity are particularly useful in determining the

effect of legal changes in PIT schedules on growth, evasion, income inequality, and other

important outcomes. Unlike structural progressivity, effective progressivity is directly derived

from collected revenues and existing income distribution, which makes the identification of the

causal effect of the effective progressivity on the outcomes from which it is derived highly

problematic. Additionally, the data requirements of our approach are far less demanding than

some of the existing approaches that calculate effective tax rates (Easterly and Rebelo 1993,

Mendoza, Razin, and Tesar 1994).23

22 Transition to simpler income tax systems does not necessarily imply the drop in the cost of compliance and tax administration. The costs may increase even if there is a reduction in complexity (Blumenthal and Slemrod 1992, Tran-Nam, Evans, and Walpole 2000). 23 Full implementation of Easterly and Rebelo’s (1993) method requires data on income distribution, statutory tax rates, and actual tax revenues collected. Assumptions about the tax function, the income distribution, and the level at which taxes are paid are also required (e.g., taxes are assumed to be paid at the household level and that each household has 5 members). Only 32 countries were included in their analysis due to data requirements. Although Mendoza et al. (1994) rely on fewer parametric assumptions, their approach also requires data that are not readily available for many countries. Detailed data on national accounts (e.g., wages, consumption, operating surplus of the

20

To assess the structural PIT progressivity for each country in the sample, we calculate the

marginal rate progression (MRP) and average rate progression (ARP). Unlike Musgrave and

Thin (1948), whose progressivity measures vary over the income distribution, we develop a

single, comprehensive measure for each country using the following procedure. First, we

compute marginal and average rates for each country and each year at 100 different levels of pre-

tax income that are evenly spread in the range from 4% to 400% of a country’s GDP per capita

(the calculation process is described earlier in the text). These variables are then used to

construct marginal and average rate progression indices (MRP1 and ARP1, respectively) by

regressing marginal (or average) rates on the log of gross income. For example, ARP1 for

Canada in 2005 is the estimated slope coefficient from the regression of average tax rates in 2005

on the log of gross income using 100 data points that are formed around Canada’s GDP per

capita. The tax structure is interpreted as progressive, proportional or regressive if the slope is

positive, zero, or negative, respectively. Thus, in order to get a single, comprehensive measure,

we had to impose a linearity restriction on the relationship between rates and income levels. In

an effort to capture some nonlinearity of these measures, we also calculate MRP2 and ARP2 for

the middle portion of income distribution in the range from 100% to 300% of a country’s GDP

per capita.

Table 7, Panel A provides the GDP-weighted means of the progressivity measures over

the sample period. We observe that both progressivity measures were positive for the full

sample period indicating that the tax structures remain progressive despite many reforms. It is

also shown that structural progressivity declined over the sample period. We note, for example,

economy, etc.) and tax revenues (for different sources and at different levels of government) are needed to calculate their effective tax rates. Besides the fact that these data are not available for many countries and years, comparability across countries remains an issue where the data are available. In fact, the authors initially calculated their tax measure for G7 countries only and later included 11 additional OECD member countries.

21

that MRP1 fell from 9.98 in 1981-1985 to 6.96 in 2001-2005, while the ARP1 fell from 8.01 in

1981-1985 to 5.46 in 2001-2005. When examined on a period basis, in Table 7, we observe a

noticeable decline in all progressivity measures throughout the late 1980s and early 1990s. This

is consistent with our earlier discussion which highlighted these years as periods of significant

tax reforms. Panel C shows that after accounting for country fixed effects and serial correlation,

the average within-country trend for progressivity measures stays negative and for the most part

statistically significant, with the exception of MRP2 that has large standard errors.

We also note that the structural progressivity of the income tax system is smaller in the

middle of the income distribution (MRP2 and ARP2) than for the broader distribution (MRP1

and ARP1), thus confirming the non-linearity of the progressivity measures.

Along with the average worldwide trend toward lower structural progressivity, we also

observe significant variation in the progressivity dynamics across country type. Panel B in Table

7 shows that countries that experienced the strongest decline in the progressivity measures

belong to the high income and upper middle income categories. The downward trend slope is

much smaller in lower middle income countries, and the trend is positive and statistically

significant in the poorest developing economies. This result is not surprising given our earlier

finding of an increase in actual tax rates at the top of the distribution in low income countries.

Thus, data analysis points to personal income tax schedules becoming increasingly flatter

over time.24 Whether these changes translated into lower or higher effective progressivity (and

associated income inequality) remains an open question. The relationship between structural

progressivity and effective progressivity is quite complex. In theory, it is possible for structural

24 This trend is not limited to the personal income tax. In fact, evidence suggests that corporate income tax schedules are also flatter now than they were in the past. The share of countries with progressive corporate income tax schedule declined from 32% in 1981 to just 12% in 2005. Furthermore, despite some modest variation across countries, consumption taxes are generally administered at a single flat rate.

22

and effective progressivity to move in opposite directions following changes in tax rates.

Musgrave and Thin (1948) showed that a shift to a higher revenue-yielding tax structure that

maintains the same level of average rate progression at all points of the income distribution will

cause effective progression to increase. Furthermore, if the tax reform increases productivity and

reduces tax evasion, then a reduction in structural progressivity may actually result in higher

effective progressivity, and hence greater equity, assuming that the redistribution of the increased

tax revenues is progressive or pro-poor. But even when redistribution from rich to poor does not

occur, lower taxes on the rich could in principle decrease true net income inequality if the share

of hidden income among the rich is large while the elasticity of true income/productivity is small

relative to the elasticity of hidden income with respect to tax changes (Duncan and Sabirianova

Peter 2008).

While we focus on personal income taxes, it is also important to acknowledge the role of

other taxes in determining the overall effective progressivity of the tax system. Changes in other

aspects of the tax system can either reinforce or offset the equity implications of the fall in PIT

progressivity that we document. For example, developing countries have increased their reliance

on consumption taxes while reducing their dependence on tariffs over the period of our study

(Bahl and Bird 2008). Because consumption taxes are known to be regressive, the net effect of

these changes on effective progressivity is likely to be ambiguous for low income countries;

recall that PIT progressivity increased for the poorest developing countries. The scientific

assessment of how reduced PIT structural progressivity may have contributed to the overall

effective progressivity requires detailed micro-level analysis, which is difficult to implement on

a large cross-country basis due to significant data requirements.

23

8. Flat Tax Revolution

The outmost case of zero (or near to zero) structural progressivity deserves special

consideration. It occurs when a country has either no tax25 or a one-rate flat tax that is not

dependent on where an individual falls in the income distribution. The case of the flat personal

income tax rate is particularly interesting as the number of countries introducing the flat tax rate

is rapidly growing in recent years.26 The flat tax rate may or may not be accompanied by an

allowance or an additional zero-rate threshold, which could make progressivity measures

different from zero.

Only 1.6% of countries had flat rate income tax schedules during the 1981-1985 period.

This number doubled to 3.67% over the next ten years and almost quadrupled to 6.23% by 2001-

2005. By 2005, fourteen countries worldwide have adopted the one-rate personal income tax,

eleven of which are middle income countries.

While still in its early stages, the trend toward flatter taxes continues to gather strength as

the list of countries adopting the flat tax expands. Since 1994, for example, when Estonia

adopted the first flat tax in Europe, Georgia, Latvia, Lithuania, Russia, Romania, Slovakia,

Ukraine, and Serbia have all adopted flat tax rate schedules. This group has increased to 21 by

2008, as seven more countries (Albania, Kazakhstan, Kyrgyzstan, Mongolia, Montenegro,

Poland, and Republic of Macedonia) introduced the flat PIT rate in 2007-2008.

Despite the increasing popularity of the flat rate, there continues to be significant debate

among countries about the merits of making the switch to a flat tax. Gorodnichenko, Martinez- 25 The number of countries with zero tax was relatively stable throughout the sample period, between 15 and 17 countries. These countries include some of the oil countries (e.g., Brunei Darussalam, Saudi Arabia, United Arab Emirates, etc.) and famous offshore islands and tax heavens (e.g., Bahamas, Bermuda, British Virgin Islands, Cayman Islands, etc.). 26 It should be noted that the flat PIT rate is not a pure flat consumption tax as outlined in Hall and Rabushka (1985). Different rates may still apply to capital gains and business income. The presence of deductions in some countries means that the ‘flat’ PIT rate may not be proportional.

24

Vazquez, and Sabirianova Peter (2009), for example, show that the Russian flat tax reform led to

a significant decline in tax evasion, but only to a small increase in productivity. This study

concludes that such reforms could be useful in countries experiencing large tax evasion but the

governments should not expect a large productivity response. It could explain why flat tax

reforms are especially popular among former socialist countries that are plagued by large shadow

economic activities. Although there is a need for more work on the equity and efficiency effects

of the flat tax, it is reasonable to conclude that the long term trend towards lower tax rates and

broader tax bases may be entering a new era of change where countries move away from their

conventional stair stepped progressive income tax schedules to simpler, flatter tax structures.

9. PIT Rates and PIT Revenue

Having examined the major trends in personal income tax rate structure, the natural

follow-up question is whether and how these changes influenced collected PIT revenue. Since

tax revenue is the product of the tax rate and tax base, determining what happened to collected

revenues implicitly tells us what happened to the tax base. Our objective in this section is 1) to

establish the trend in PIT revenue over the sample period, and 2) to examine the relationship

between PIT revenue and our tax measures.

Existing, but limited, data show almost no change in the share of PIT revenue in GDP

over time. From Panel A of Figure 5, we observe that the total PIT revenue as a share of GDP

for OECD countries has been trending slightly downward but stayed in the range of 9% to 11%

throughout the 1981-2005 period. A similar pattern is observed in Panel B of Figure 5 where we

report the trend in the PIT revenue of the central government as a share of GDP for both

25

developed and developing countries.27 High income countries display a slight downward trend

in the PIT/GDP ratio, which stays in the range of 6% to 8%. This is about two thirds of the PIT

revenue share for the general government that includes both central and local/state/province

governments. In developing countries, the central government PIT share in GDP does not have a

visible trend and has an average range of approximately 1 percentage point for each of the upper

middle, lower middle and low income country groups.28 Similar to previous studies, we find that

the PIT/GDP ratio is significantly higher in more developed countries.

A more rigorous analysis of these trends is provided in Table 8 where we report the trend

coefficients for PIT and total income tax revenue to GDP ratios. While the pooled OLS results

show a statistically significant decline over the sample period, we find no evidence of a trend in

both ratios when we include country fixed effects. Given the overall decline in average PIT rates

described above, one implication of these results is that the PIT tax base increased over time.

However, there appears to be some differences across country groups for PIT revenue (but not

for total income tax revenue). In particular, we find evidence of a statistically significant within-

country decline in the PIT/GDP ratio for high income countries (the coefficient on Trend in

column 3). Lower middle and low income countries, on the other hand, have a modest upward

trend after controlling for country fixed effects. These results are very interesting in light of the

changes in PIT rates described in sections 4 through 7. In fact, the trends seem to imply that the

27 The data used in Panel A are from OECD.Stat for about 25-29 countries per year. Panel B uses data from the International Monetary Fund’s Government Financial Statistics for non-OECD countries and OECD.Stat for OECD countries for 59-80 countries per year. 28 The ranges are 2%-3%, 1.7%-2.8%, and 0.7%-2% for upper middle, lower middle and low income countries, respectively. These results are consistent with findings elsewhere in the literature. For example, Bird and Zolt (2005), looking at the period 1981 to 1992 for OECD, African, and Asian countries, find similar evidence; personal income tax as a share of GDP did not change much. Using IMF Government Finance Statistics for 28 developing countries, Alm and Wallace (2006) report a small increase in PIT collections from 1.9% of GDP in 1986-1988 to 2.5% of GDP in 1996-1998.

26

lower PIT rates are associated with lower PIT revenue in high income countries and with higher

PIT revenue in low income countries.

We explore this finding further by regressing the PIT/GDP ratio on selected tax variables,

including the top rate, average and marginal rates, and PIT progressivity measures. The results

with country fixed effects are reported in Table 9. Since PIT collection may influence which tax

rates countries choose, the endogeneity of tax rates does not allow us to make any causal

interpretations. Still, the estimates of within-country correlation between tax variables and PIT

collection are quite informative. Panel A shows that there is a positive and statistically

significant relationship between all our tax variables and the PIT revenue share. More

interesting are the results in Panel B where we interact the tax measures with country category to

see how the relationship varies with the level of economic development. The estimates imply

that the PIT revenue share and tax rates move in the same direction in developed countries.

However, the interaction terms of tax variables with other country categories are all negative

though not always statistically significant when standard errors are clustered by country. For

poor countries, we find that the interaction effect cancels out the main effect of the top PIT rate,

marginal tax rate at income equivalent to four times GDP per capita, and both progressivity

measures (MRP1 and ARP1). That is, the PIT revenue share is not found to be positively

associated with higher marginal rates and progressivity measures in low income countries.

These results do raise additional questions. The most obvious being why is there such

difference in the relationship between PIT rates and PIT revenue across country types? The

following decomposition may help to answer to this question. Suppose R/Y is the share of PIT

revenue in GDP, where revenue, R, is the product of the PIT base, B, and the average PIT rate, τ.

Then, the response of the PIT/GDP ratio to changes in the tax rate can be written as:

27

( ) ( )( ) ( )⎟⎠⎞

⎜⎝⎛ ∈−∈+=⎟⎟

⎠

⎞⎜⎜⎝

⎛∂∂

−⎟⎠⎞

⎜⎝⎛

∂∂

+=∂⋅∂

=∂

∂ −− )()(

2 11//τττ

ττ

ττ

ττττ YBY

BYBBBYY

YBYR , (1)

where τB∈ and τY∈ are the elasticities of the PIT base and GDP in response to the tax rate,

respectively. Equation (1) implies that the larger is the PIT base relative to GDP, the stronger is

the response of PIT revenue to tax changes. Therefore, one possible explanation for the

observed weak relationship between the tax rate and PIT revenue in developing countries is that

their PIT base is small relative to their GDP. This, in turn, could be due to legal rules restricting

taxable income, but more likely due to significant informal sector activities. Given that the tax

base, by definition, does not include unreported income while GDP often accounts imperfectly

for some portions of the informal sector (especially since adopting the 1993 System of National

Accounts), the ratio of the tax base to GDP is likely to be inversely related to the size of the

shadow economy.

However, we should not discount the importance of the second term in equation (1),

( )ττ YB ∈−∈+1 . The difference in the responsiveness of the tax base relative to the

responsiveness of GDP is another possible reason for the cross-country differences observed in

Tables 8 and 9. In other words, tax induced behavioral responses may vary across countries in

systematic ways that then lead to the observed trends.

Conventional wisdom in the taxation literature is that timing and accounting-form

responses of the tax base are generally larger than real responses (Slemrod 1990). This is

supported by recent evidence that the productivity response to the Russian flat tax reform was

small relative to the compliance response (Gorodnichenko, Martinez-Vazquez, and Sabirianova

Peter 2009). If the tax base contracts more sizably relative to GDP in reaction to higher tax rates

due to a large evasion response, as is probably the case in countries with substantial informal

28

economies, then the absolute difference between the elasticities in equation (1) becomes positive,

0>∈−∈ ττ YB , thus reducing the responsiveness of the PIT/GDP ratio to tax rates. Again, we

see the link between the large size of the shadow economy and the weak response of the PIT

revenue share to tax changes.

While we cannot perform the formal testing of this hypothesis, we can check if the

responsiveness of the PIT revenue share to tax rates declines with weaker institutions and, in

particular, with the level of corruption that is likely to be correlated with the size of the informal

sector. The results, reported in Panel C of Table 9, show that the relationship between the tax

rate and PIT/GDP ratio is positive and statistically significant for countries with very low

corruption, but the relationship significantly weakens with higher levels of corruption. These

results imply that the size of the informal sector may have contributed to the observed weak

relationship between the PIT revenue and the tax rate in developing countries.

10. Importance of Large Representative Samples

We pointed out earlier that one of the contributions of our study is the extensive coverage

we provide across time and countries. Most existing studies looking at the effect of taxes on

economic outcomes tend to focus on a small group of countries (e.g., OECD countries) and thus

may suffer from sample selection issues. Our study addresses these issues by providing a dataset

that is both representative and highly comparable. In this section we provide a simple analysis

that highlights the importance of having a large representative sample. We also show that the

changing composition of our sample over time has no effect on the trends discussed so far.

Year to year changes in the number of countries are unavoidable in long panels like ours.

These changes are mostly due to 1) the lack of data in early years for some countries, and 2) the

formation of new countries and the dismantling of old ones. The former mainly affects

29

developing countries while the latter applies primarily to transitional countries that emerged

following the breakup of communism in the early 1990’s.29 The number of countries per year in

our sample starts at 108 in 1981 and steadily increases up to a max of 180 in 2004, with the total

number of covered countries being 189. Even at 108 countries, our initial sample is highly

representative and accounts for approximately 86% of world GDP. The 1981 sample also

represents every geographical region of world, and all World Bank income groups.30

Nonetheless, it remains informative to examine the impact, if any, of these changes on the trends

over time. We test for any such effects by estimating the trend coefficient on tax variables using

only the initial set of countries and comparing the results with those obtained using the full

sample. Any compositional effect should show up as the difference between these two trend

coefficients. The results in Table 10 show that there is no statistically significant difference at

the 5% level in the estimated trend coefficients and hence no compositional effect.31 These

findings imply that we can be confident that the results we document in the previous sections are

not influenced in any significant way by changes in the number of countries over time.

In addition to the compositional effect, we also emphasize the importance of having a

large representative panel of countries. Sabirianova Peter (2008) demonstrates that using panel

data methods on a sample of 170+ countries switches the sign of the effect of various tax

variables on the size of the shadow economy from negative in OLS to positive and highly

29 We gathered tax information on countries before and after they breakup to form new countries. For example, our dataset includes information for the USSR from 1983 up to its breakup in 1991. Data is then collected for all 15 new countries formed after the breakup for the remainder of our sample. Other countries that went through breakups are treated similarly, e.g., Czechoslovakia, Yugoslavia, and Ethiopia (with Eritrea included). 30 The 1981 sample includes 19 African, 22 Asian/Oceanic, 29 Central/South American and Caribbean, 20 European, 9 Mid East and 4 North American (includes Bermuda) countries. Additionally, we have 31, 18, 38 and 15 High, Upper Middle, Lower Middle and Low Income countries, respectively, in the 1981 sample. 31 We find statistically significant differences between the trend coefficients at the 10% level only for MRP1 in OLS estimations. In all other cases (including all FE estimations), the trend slopes estimated on the 1981 sample (108 countries) are not statistically significant at the 10% level from the trend slopes estimated when all countries are included. The results for other tax variables are similar and not reported.

30

statistically significant in fixed effect models. In this paper, we illustrate the importance of large

representative samples in the context of a simple GDP growth model.32 We estimate the effect

of the top PIT rate on the growth of GDP using several alternative data sources (University of

Michigan WTD, Heritage Foundation, OECD, World Bank, and the Economist Intelligence

Unit) and compare the estimates with the estimates from our data (World Tax Indicators). The

results from this exercise are reported in Table 11. We first note that alternative data sources are

smaller in size and typically under-represent low-income countries (see Panel A). Panel B

reports the trend coefficient on the top PIT rate and shows the large variation in sign, size and

significance by data source. For example, Heritage Foundation and Economist Intelligence Unit

databases show no trend in the top PIT rate, OECD and World Bank data sources produce a large

downward bias in the estimated trend while the WTD shows a relatively steep descending trend

in top rates. From Panel C, we see that only our dataset shows a negative and statistically

significant relationship between the top PIT rate and GDP growth rate. The estimates of the tax

rate effect on growth from other data sources are all statistically insignificant.

These results make it clear that reliance on small unrepresentative samples may lead to

incorrect inferences and thus affect policy prescriptions. In this respect, we believe that our

dataset is superior to most of the existing data on PIT rates.

11. Conclusion

This paper introduces a panel dataset on the PIT structure that is highly comparable

across countries and years, and unprecedented in its coverage. The dataset includes the measures

32 The growth model includes the top PIT rate, annual change in the share of government consumption expenditures in GDP, annual change in the share of gross fixed capital formation in GDP, population growth rate, the share of industrial production in value added, year dummies, and country fixed effects to control for a country’s endowment in human and physical capital and other time-invariant characteristics. We use the top PIT rate for comparability with other data sources.

31

of structural progressivity as well as average and marginal PIT rates at different points of the

income scale. Also included is information on tax brackets, statutory rates, basic allowances, tax

credits, etc. These features make our data the desirable source for studies of the relationship

between tax policy and economic outcomes in a country level setting.

Using this dataset, we identify and determine the direction of a momentous change in the

world’s tax systems. We highlight a tax reform process in which governments have moved away

from complex, progressive tax systems featuring multiple tax brackets and escalating stair step

tax rates to simpler, flatter tax schedules distinguished by fewer tax brackets and lower tax rates.

In fact, taxpayers on average face lower personal income tax rates and simpler rate structures

than they did twenty five years ago. This transformation suggests an epic and ubiquitous change

in fiscal policy.

As the trends continue to play out, we also see a nascent secondary trend emerging where

the conventional stair-stepped tax schedule is being replaced by a flat tax rate. The recent

acceleration in this trend began with Estonia’s adoption of a flat tax in 1994 and has since spread

to other post-communist countries. As noted earlier, the number of countries using flat tax rates

has soared since 1981 but still remains relatively small at about 12% of the sample countries by

2007. However, at the writing of this paper, other countries such as Croatia, Czech Republic,

Greece, and Mexico are giving a flat tax schedule serious consideration. It seems to us that the

trend toward flat taxes is likely to continue for the next several years. Whether or not this second

trend continues, exceptional long term changes have already transformed the tax systems of the

world.

We also analyzed the relationship between PIT rates and PIT revenue collection and find

that the relationship varies in important ways across countries. In particular, we find a

32

statistically significant positive association between tax rates and PIT collection in high income

countries. Developing countries, on the other hand, show very weak relationship between tax

rates and the PIT/GDP ratio. Our results imply that the size of the informal sector may have

contributed to the observed weak relationship between the PIT revenue and the tax rate in

developing countries.

The observed trends raise a number of questions that we briefly addressed in the current

paper and hope to address more carefully in future work. For example, what are the implications

for efficiency, equity, and growth? Conceivably, flatter tax structures could reduce the excess

burden without necessarily sacrificing equity. Davies and Hoy (2002), for example, argue that

moving from a graduated tax schedule to a flat tax may improve equity. Furthermore, flatter,

simpler tax systems can possibly increase the efficiency of the tax system by broadening the tax

base as well as improving tax compliance.33 On the other hand, flatter tax systems can lead to

less equity as more of the tax burden is redistributed from the wealthier taxpayers to the lower

and middle income classes. The effect of these changes on growth is also of grave importance.

To the extent that efficiency has increased, we might expect a positive effect on growth. If

inequality also increased, however, then the effect on growth may be ambiguous. That is,

increased inequality arising from the tax reform may or may not lead to increased growth

(Milanovic 2000, Persson and Tabellini 1994, Perotti 1992, among others). As such, the net

effect on equity, efficiency, and growth remains an empirical question.

33 Gorodnichenko, Martinez-Vazquez, and Sabirianova Peter (2009). Feldstein (1995) demonstrated that the sensitivity of taxable income with respect to changes in the marginal tax rate is much larger than can be explained by estimated labor supply elasticities. The implication then is that there are other behavioral responses at work.

33

References

Alm, J., and S. Wallace, 2006. "Can Developing Countries Impose an Individual Income Tax?," in J. Alm, J. Martinez-Vazquez, and M. Rider, editors, The Challenges of Tax Reform in a Global Economy: New York: Springer, 221-248.

Bahl, R. W., and R. M. Bird, 2008. "Tax Policy in Developing Countries: Looking Back--and Forward," National Tax Journal, 61(2): 279-301.

Bird, R. M., and E. M. Zolt, 2005. "The Limited Role of the Personal Income Tax in Developing Countries," Journal of Asian Economics, 16(6): 928-46.

Blumenthal, M., and J. Slemrod, 1992. "The Compliance Cost of the U.S. Individual Income Tax System: A Second Look after Tax Reform," National Tax Journal, 45(2): 185-202.

Davies, J. B., and M. Hoy, 2002. "Flat Rate Taxes and Inequality Measurement," Journal of Public Economics, 84(1): 33-46.

Duncan, D., and K. Sabirianova Peter, 2008. “Tax Progressivity and Income Inequality,” Andrew Young School of Policy Studies Research Paper Series, No. 08-26.

Easterly, W., and S. Rebelo, 1993. "Marginal Income Tax Rates and Economic Growth in Developing Countries," European Economic Review, 37 (2-3): 409-17.

Feldstein, M., 1995. "The Effect of Marginal Tax Rates on Taxable Income: A Panel Study of the 1986 Tax Reform Act," Journal of Political Economy, 103 (3): 551-72.

Friedman, E., S.Johnson, D.Kaufman and P.Zoido-Lobaton, 2000. "Dodging the Grabbing Hand: The Determinants of Unofficial Activity in 69 Countries," Journal of Public Economics, 76(3): 459-493.

Gorodnichenko, Y., J. Martinez-Vazquez, and K. Sabirianova Peter, 2008. "Myth and Reality of Flat Tax Reform: Micro Estimates of Tax Evasion Response and Welfare Effects in Russia," Journal of Political Economy, forthcoming.

Hall, R. E., and A. Rabushka, 1985. The Flat Tax. Stanford: Hoover Institution Press.

Johnson, S., D. Kaufmann, and P. Zoido-Lobaton, 1998. "Regulatory Discretion and the Unofficial Economy," American Economic Review, 88(2): 387-392.

Lee, Y., and R. H. Gordon, 2005. "Tax Structure and Economic Growth," Journal of Public Economics, 89(5-6): 1027-1043.

Mendoza, E. G., A. Razin, and L. L. Tesar, 1994. "Effective Tax Rates in Macroeconomics: Cross-Country Estimates of Tax Rates on Factor Incomes and Consumption," Journal of Monetary Economics, 34 (3): 297-323.

34

Milanovic, B., 2000. "The Median-Voter Hypothesis, Income Inequality, and Income Redistribution: An Empirical Test with the Required Data," European Journal of Political Economy, 16 (3): 367-410.

Musgrave, R., and T. Thin, 1948. "Income Tax Progression, 1929-48," Journal of Political Economy, 56(6): 498-514.

Paulus, A., and A. Peichl, 2008. "Effects of Flat Tax Reforms in Western Europe on Income Distribution and Work Incentives," The Institute for the Study of Labor (IZA) Discussion Paper, No. 3721.

Perotti, R., 1992. "Income Distribution, Politics, and Growth," American Economic Review, 82 (2): 311-16.

Persson, T., and G. Tabellini, 1994. "Is Inequality Harmful for Growth?," American Economic Review, 84( 3): 600-621.

Sabirianova Peter, K., 2008 " Income Tax Flattening: Does It Help to Reduce the Shadow Economy?" Andrew Young School of Policy Studies Research Paper Series, No. 08-09.

Seligman, E.R.A., 1908. “Progressive Taxation in Theory and Practice,” American Economic Association Quarterly, 9(4): 1-334.

Sicat, G. P., and A. Vermani, 1988. "Personal Income Taxes in Developing Countries," World Bank Economic Review, 2(1): 123-138.

Slemrod J., 1990. Do Taxes Matter? The Impact of the Tax Reform Act of 1986. Cambridge, Mass. and London: MIT Press.

Slemrod, J., 1992. "Did the Tax Reform Act of 1986 Simplify Tax Matters?" Journal of Economic Perspectives, 6 (1): 45-57.

Slemrod, J., and N. Sorum, 1984. "The Compliance Cost of the U.S. Individual Income Tax System," National Tax Journal, 37(4): 461-474.

Tran-Nam, B., C. Evans, and M. Walpole, 2000. "Tax Compliance Costs: Research Methodology and Empirical Evidence from Australia," National Tax Journal, 53(2): 229-252.

35

Table 1: Frequency of Changes in PIT Schedule by Period and Country Category

Period High Income Upper Middle Income

Lower Middle Income

Low Income Total

Panel A: Share of countries with no yearly changes in statutory rates and thresholds

1981 – 1985 0.348 0.667 0.735 0.661 0.592 1986 – 1990 0.298 0.560 0.623 0.671 0.526 1991 – 1995 0.419 0.525 0.500 0.606 0.506 1996 – 2000 0.439 0.491 0.545 0.747 0.558 2001 – 2005 0.378 0.576 0.571 0.782 0.574 All years 0.383 0.552 0.583 0.711 0.550

Panel B: Share of countries with yearly changes in both statutory rates and thresholds

1981 – 1985 0.359 0.117 0.173 0.220 0.227 1986 – 1990 0.427 0.307 0.239 0.224 0.303 1991 – 1995 0.199 0.273 0.259 0.257 0.245 1996 – 2000 0.269 0.202 0.175 0.139 0.198 2001 – 2005 0.311 0.196 0.186 0.129 0.210 All years 0.306 0.223 0.210 0.179 0.233

Panel C: Share of countries with yearly adjusted thresholds but unchanged statutory rates

1981 – 1985 0.250 0.200 0.071 0.102 0.155 1986 – 1990 0.258 0.120 0.130 0.059 0.152 1991 – 1995 0.331 0.152 0.209 0.101 0.207 1996 – 2000 0.269 0.254 0.245 0.101 0.215 2001 – 2005 0.272 0.207 0.229 0.082 0.196 All years 0.277 0.191 0.185 0.090 0.190 N 703 440 677 581 2401 Notes: Frequencies are computed on the sample of countries with a complete national PIT schedule in both current and previous years.

36

Table 2: Distribution of Countries by Top PIT Rates

Period 0% 1-20% 21-30% 31-40% 41-60% 61-95% 1981 – 1985 13.11 5.42 2.80 6.99 46.33 25.35 1986 – 1990 12.87 8.47 5.70 14.66 47.07 11.24 1991 – 1995 11.55 10.05 14.13 32.20 29.62 2.45 1996 – 2000 10.32 11.53 19.78 34.47 22.94 0.97 2001 – 2005 9.69 15.80 23.30 34.03 16.26 0.92

Note: N=3613.

37

Table 3: Top Statutory PIT Rates by Period and Country Category

Period High Income Upper Middle Income

Lower Middle Income

Low Income Total

Panel A: GDP-Weighted Top PIT Rates (Mean and Standard Deviation)

1981 – 1985 58.012 61.111 54.188 52.138 57.936 (13.588) (11.506) (10.833) (9.961) (13.205) 1986 – 1990 46.494 55.764 48.736 47.085 47.848 (15.121) (14.115) (11.648) (6.559) (14.853) 1991 – 1995 42.663 37.546 42.242 43.312 42.179 (11.347) (15.018) (14.118) (6.737) (11.823) 1996 – 2000 42.791 33.206 39.933 36.258 41.456 (9.500) (11.768) (8.731) (8.582) (10.037) 2001 – 2005 38.908 29.546 35.818 31.823 37.739 (8.327) (11.549) (10.647) (7.001) (9.149)

Panel B: GDP-Weighted Average Trend

1981 – 1985 -1.481* 1.149 -0.367 -1.871* -1.110** [189] [99] [197] [87] [572] 1986 – 1990 -3.744** -3.445** -1.340* 0.238 -3.408** [205] [108] [202] [99] [614] 1991 – 1995 0.760 -3.313** -3.220** -1.332** 0.099 [209] [139] [242] [146] [736] 1996 – 2000 -0.870* 0.241 0.302 -2.528** -0.814**

[240] [149] [208] [227] [824] 2001 – 2005 -0.862* -0.590 0.182 -0.467 -0.745**

[261] [142] [219] [245] [867] All years -0.797** -1.752** -0.855** -1.059** -0.892**

Panel C: Within-Country Trend, FE

All years -0.551** -0.869** -1.079** -1.294** -0.876** (0.079) (0.163) (0.115) (0.199) (0.063)

[1104] [637] [1068] [804] [3613] Notes: Panel A reports means and standard deviations (in parentheses) of the top PIT rates weighted by GDP in constant 1990 U.S. dollars. Panels B and C report estimated slopes from regressing top PIT rates on the trend variable. In Panel B, the estimates are weighted by GDP in constant 1990 U.S. dollars. In Panel C, the estimates are obtained with country fixed effects and heteroskedasticity-and-autocorrelation-robust standard errors (reported in parentheses); ** significant at 1% level, * significant at 5% level. Number of observations is in square brackets, and it is the same in all three panels.

38

Table 4: Trend Coefficients for Actual PIT Rates by Country Category

Period High Income

Upper Middle Income

Lower Middle Income

Low Income Total

Panel A: GDP-Weighted Average Trend

Marginal PIT rate 1⋅y -0.491** -0.172 -0.347** 0.054 -0.491** 2⋅y -0.623** -0.309** -0.291** 0.037 -0.622** 3⋅y -0.731** -0.413** -0.378** 0.335** -0.716** 4⋅y -0.796** -0.428** -0.414** 0.716** -0.756**

Average PIT rate 1⋅y -0.329** -0.022 -0.255** 0.037 -0.323** 2⋅y -0.431** -0.128* -0.280** 0.040 -0.431** 3⋅y -0.523** -0.218** -0.296** 0.068 -0.520** 4⋅y -0.584** -0.274** -0.317** 0.214** -0.573**

Panel B: Unweighted Within-Country Trend, FE

Marginal PIT rate 1⋅y -0.404** -0.204** -0.131 -0.063 -0.218** 2⋅y -0.423** -0.291** -0.084 -0.198 -0.242** 3⋅y -0.430** -0.388** -0.084 -0.158 -0.263** 4⋅y -0.465** -0.414** -0.154 -0.147 -0.307**

Average PIT rate 1⋅y -0.272** -0.137** -0.067* -0.038 -0.146** 2⋅y -0.341** -0.181** -0.094* -0.079 -0.183** 3⋅y -0.372** -0.244** -0.085 -0.115 -0.207** 4⋅y -0.390** -0.282** -0.095 -0.121 -0.226**

[1035] [571] [827] [649] [3082] Notes: Reported are the estimated slopes from regressing marginal and average PIT rates on the trend variable; y is a country’s GDP per capita. In Panel A, the estimates are weighted by GDP in constant 1990 U.S. dollars. In Panel B, the estimates are obtained with country fixed effects and heteroskedasticity-and-autocorrelation-robust standard errors; ** significant at 1% level, * significant at 5% level. Number of observations is in square brackets, and it is the same in all panels.

39

Table 5: Complexity Components of PIT Systems by Period Allowances/credits Surtax Local Taxes Multiple Tax Standard Non -

Standard Low High Schedules Formula

Panel A: Complexity Measures by Period (Mean and Standard Deviation)

1981 – 1985 0.275 0.527 0.281 0.067 0.093 0.207 0.193 (0.447) (0.500) (0.450) (0.250) (0.290) (0.406) (0.395) 1986 – 1990 0.289 0.482 0.297 0.081 0.092 0.228 0.182 (0.453) (0.500) (0.457) (0.272) (0.290) (0.420) (0.387) 1991 – 1995 0.307 0.413 0.266 0.087 0.080 0.198 0.125 (0.461) (0.493) (0.442) (0.282) (0.271) (0.399) (0.331) 1996 – 2000 0.315 0.404 0.230 0.108 0.067 0.192 0.107 (0.465) (0.491) (0.421) (0.311) (0.250) (0.394) (0.309) 2001 – 2005 0.307 0.397 0.212 0.102 0.069 0.172 0.092 (0.462) (0.490) (0.409) (0.302) (0.253) (0.378) (0.289)

Panel B: Probit Estimates

Trend 0.001 -0.006** -0.004** 0.001 -0.002* -0.003** -0.005** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Upper middle income 0.248** -0.135** 0.049* 0.039** -0.253** 0.027 -0.148** (0.025) (0.025) (0.022) (0.013) (0.019) (0.020) (0.017) Lower middle income 0.387** -0.330** 0.037 -0.057** … -0.040* -0.146** (0.021) (0.023) (0.019) (0.014) (0.018) (0.015) Low income 0.417** -0.446** 0.038 0.061** … 0.058** -0.091** (0.023) (0.026) (0.021) (0.012) (0.019) (0.015)

Panel C: Within-Country Trend, FE

Trend -0.000 -0.006** -0.002 0.002 -0.000 -0.002 -0.003** (0.002) (0.002) (0.002) (0.001) (0.001) (0.002) (0.001) N (observations) 3388 3388 3395 3392 3392 3387 3143 N (countries) 182 182 182 182 182 182 177 Notes: Panel A reports unweighted, unconditional means with standard deviations in parentheses. Omitted categories are countries with no basic allowances or tax credits and no local taxes. Panel B reports the results of the probit regression of each complexity indicator on the trend variable and country categories, with heteroskedasticity-robust standard errors in parentheses; ** significant at 1% level, * significant at 5% level. The omitted category is high income countries. Panel C reports the fixed effect linear probability regression of each complexity indicator on the trend variable, with country-clustered robust standard errors in parentheses. Number of observations is the same in all panels.

40

Table 6: Distribution of Countries by the Number of Brackets, % Period 0 1-3 4-5 6-9 10-19 20+ 1981 – 1985 15.76 6.30 2.73 25.00 41.18 9.03 1986 – 1990 14.71 12.10 11.92 25.70 30.17 5.40 1991 – 1995 13.30 19.41 26.60 26.29 12.68 1.72 1996 – 2000 11.50 21.65 28.28 28.01 9.74 0.81 2001 – 2005 11.17 23.54 31.78 27.93 5.19 0.40 Note: N=3613. The zero bracket category applies to countries that do not have a PIT.

41

Table 7: Structural Progressivity Measures of PIT Systems by Period

Period Marginal Rate Progression Average Rate Progression N MRP1 MRP2 ARP1 ARP2

Panel A: GDP-Weighted Progressivity Measures

1981 – 1985 9.982 8.912 8.011 7.012 459 (3.563) (4.090) (2.877) (2.519) 1986 – 1990 8.331 7.177 6.906 6.090 517 (3.324) (4.129) (2.514) (2.288) 1991 – 1995 7.521 7.056 5.909 5.215 624 (3.140) (4.129) (2.381) (2.181) 1996 – 2000 7.220 5.986 5.788 5.141 734 (2.603) (2.958) (2.376) (2.074) 2001 – 2005 6.963 5.901 5.461 5.006 748 (2.506) (3.222) (2.293) (2.100)

Panel B: GDP-Weighted Average Trend

All countries -0.133** -0.135** -0.117** -0.092** 3082 High income -0.141** -0.145** -0.116** -0.090** 1035 Upper middle income -0.141** -0.162** -0.093** -0.095** 571 Lower middle income -0.040** -0.017 -0.043** -0.012 827 Low income 0.194** 0.090** 0.041** 0.007 649

Panel C: Unweighted Within-Country Trend, FE

All countries -0.045** -0.031 -0.041** -0.028* 3082 (0.015) (0.022) (0.010) (0.011) Notes: MRP1 and ARP1 are marginal and average tax rate progressions up to an income level equivalent to four times y, where y is a country’s GDP per capita; MRP2 and ARP2 are marginal and average tax rate progressions for the levels of income between y and 3⋅y. Means and standard deviations (in parentheses) of progressivity measures in Panel A are weighted by GDP in constant 1990 U.S. dollars. Panels B and C report estimated slopes from regressing progressivity measures on the trend variable. In Panel B, the estimates are weighted by GDP in constant 1990 U.S. dollars. In Panel C, the estimates are obtained with country fixed effects and heteroskedasticity-and-autocorrelation-robust standard errors (reported in parentheses); ** significant at 1% level, * significant at 5% level.

42

Table 8: Income Tax Revenue by Country Type

PIT/GDP All Income Taxes/GDP OLS FE FE OLS FE FE

Trend -0.061*** -0.010 -0.052* -0.088*** -0.026 -0.026 (0.022) (0.017) (0.030) (0.029) (0.025) (0.030) Upper middle income -3.692*** -0.583** -1.111* -2.682** -1.329*** -0.721 (0.850) (0.283) (0.578) (1.057) (0.444) (0.958) Lower middle income -4.120*** -0.781* -1.804** -3.897*** -1.695*** -2.411** (0.799) (0.417) (0.696) (0.980) (0.593) (1.006) Low income -4.985*** -1.402** -2.608*** -5.050*** -1.750** -2.569 (0.799) (0.616) (0.891) (1.063) (0.773) (1.596) Upper middle income … … 0.034 … … -0.066 × Trend (0.039) (0.077) Lower middle income … … 0.079** … … 0.027 × Trend (0.036) (0.044) Low income … … 0.091** … … 0.032 × Trend (0.037) (0.059) Cash vs. accrual -1.031* 0.371 0.061 -1.828** -0.610 -0.646* (0.610) (0.274) (0.291) (0.773) (0.419) (0.383) Intercept 8.105*** … … 11.532*** … … (0.927) (1.134) N (observations) 1695 1695 1695 1956 1956 1956 N (countries) 118 118 118 126 126 126 R-squared 0.36 0.03 0.07 0.23 0.02 0.04

Notes: Country-clustered robust standard errors are in parentheses; *** significant at 1% level, ** significant at 5% level, * significant at 10% level. Table reports the results of unweighted regression of income tax revenue as percent of GDP on the time trend, country type, and their interaction, controlling for the reporting revenue base (cash or accrual). The omitted category is high income countries. FE = estimator with country fixed effects. PIT = personal income tax. All income taxes are taxes on income, profits, and capital gains payable by individuals and corporations.

43

Table 9: PIT Rates and PIT Revenue

Top PIT rate

MTR at 4⋅y

MTR at 2⋅y

ATR at 4⋅y

ATR at 2⋅y MRP1 ARP1

Panel A: FE Estimates

Tax variable 0.018** 0.043*** 0.042*** 0.057*** 0.055*** 3.432* 12.897*** (0.009) (0.010) (0.010) (0.011) (0.013) (1.782) (4.483) R-squared 0.04 0.11 0.09 0.10 0.08 0.04 0.06

Panel B: FE Estimates with Country Type Interactions

Tax variable 0.047** 0.061*** 0.052*** 0.068*** 0.061*** 6.767* 19.537*** (0.019) (0.012) (0.011) (0.012) (0.014) (3.484) (5.884) Upper middle income -0.033* -0.024** -0.024 -0.025 -0.033 -4.597 -6.514 × Tax variable (0.018) (0.012) (0.016) (0.017) (0.023) (3.752) (5.785) Lower middle income -0.040* -0.032* -0.028 -0.033 -0.033 -4.884 -12.043* × Tax variable (0.021) (0.017) (0.018) (0.020) (0.027) (4.337) (6.854) Low income -0.042* -0.052*** -0.037 -0.038 0.003 -11.265** -21.388** × Tax variable (0.021) (0.018) (0.023) (0.036) (0.043) (4.431) (9.275) R-squared 0.07 0.12 0.10 0.11 0.09 0.05 0.07 N (observations) 1500 1379 1379 1379 1379 1379 1379 N (countries) 103 97 97 97 97 97 97

Panel C: FE Estimates with Corruption Interactions

Tax variable 0.051*** 0.061*** 0.057*** 0.072*** 0.069*** 4.356 20.909*** (0.019) (0.015) (0.015) (0.017) (0.018) (4.357) (6.012) Corruption index 0.297 0.146 0.048 0.050 -0.032 -0.206 -0.000 (0.256) (0.169) (0.158) (0.162) (0.146) (0.127) (0.152) Corruption index -0.014** -0.013** -0.013** -0.015** -0.016* -0.705 -5.550** × Tax variable (0.006) (0.005) (0.006) (0.007) (0.009) (1.584) (2.587) R-squared 0.09 0.13 0.11 0.12 0.11 0.05 0.07 N (observations) 1244 1165 1165 1165 1165 1165 1165 N (countries) 92 87 87 87 87 87 87

Notes: Country-clustered robust standard errors are in parentheses; *** significant at 1% level, ** significant at 5% level, significant at 10% level. Table reports the results of unweighted regression of PIT revenue as percent of GDP on the time trend, country type, and variables shown in the table, controlling for the reporting revenue base (cash or accrual). The omitted category is high income countries. All estimates include country fixed effects. MTR = marginal PIT rate; ATR = average PIT rate; MRP1 and ARP1 are marginal and average tax rate progressions up to an income level equivalent to four times y, where y is a country’s GDP per capita. The number of observations is the same in Panels A and B. Corruption index is taken from the International Country Risk Guide (ICRG) but it is rescaled from 1 to 6, where 6 indicates worse corruption.

44

Table 10: Sensitivity of Trend Estimates to Changing Composition of Countries

OLS FE All 1981 All 1981 Countries Sample Countries Sample Top PIT rate -0.796** -0.867** -0.876** -0.849** (0.080) (0.067) (0.063) (0.068) [189/3613] [108/2612] [189/3613] [108/2612] Chi2 test for difference between coefficients p-value=0.128 p-value=0.322

Marginal PIT rate at 4⋅y -0.308** -0.289** -0.307** -0.299** (0.087) (0.074) (0.060) (0.066) [175/3082] [105/2353] [175/3082] [105/2353] Chi2 test for difference between coefficients p-value=0.667 p-value=0.621

MRP1 × 100 -0.055** -0.037** -0.045** -0.042** (0.020) (0.019) (0.015) (0.016) [175/3082] [105/2353] [175/3082] [105/2353] Chi2 test for difference between coefficients p-value=0.088 p-value=0.540

Notes: Country-clustered robust standard errors are in parentheses; ** significant at 1% level, * significant at 5% level. Number of countries/observations is reported in brackets. Table reports the unweighted estimates of the slope coefficient from regressing selected tax variables on the trend variable using the sample indicated at the top of each column. The 1981 sample includes countries with available data in 1981.

45

Table 11: Sensitivity of Estimates to Different Data Sources

World Tax Indicators

Univ. Michigan

WTD

Heritage Foundation OECD

World Bank WDI

Economist Intelligence

Unit Panel A: Composition of Countries, %

Low income 15.31 12.26 13.78 0.00 8.41 1.53 Lower middle 25.05 25.11 20.51 0.00 17.29 14.80 Upper middle 24.63 25.58 27.88 6.78 28.04 23.47 High income 35.01 37.05 37.82 93.22 46.26 60.20 N (observations) 3613 1892 312 177 647 196

Panel B: Trend Coefficient for Top PIT Rate

Trend -0.876** -1.220** -0.528 -0.658** -0.520** 0.159 (0.063) (0.099) (0.496) (0.134) (0.114) (0.495) N (observations) 3613 1892 312 177 647 196 N (countries) 189 134 158 30 127 58 N (years) 25 23 2 6 6 4

Panel C: Effect of Top PIT Rate on GDP Growth

Top PIT rate -0.076** -0.016 -0.016 0.076 -0.060 0.016 (0.025) (0.021) (0.067) (0.064) (0.052) (0.048) N (observations) 3018 1619 297 177 588 196 N (countries) 168 119 152 30 114 58 N (years) 24 22 2 6 6 4

Note: Country-clustered robust standard errors in parentheses; ** significant at 1% level, * significant at 5% level. Panel B reports the unweighted estimates of the slope coefficient from regressing top PIT rates on the trend variable using the data sources indicated at the top of each column. Estimation is done by OLS with country fixed effects. Panel C presents the estimates of the relationship between the top PIT rate and real GDP growth rate using data from different sources. The growth model controls for the annual change in the share of government consumption expenditures in GDP, the annual change in the share of gross fixed capital formation in GDP, the population growth rate, the share of industrial production in value added, and year dummies.

46

2030

4050

6070

%

1980 1985 1990 1995 2000 2005

Unweighted GDP-weighted

Figure 1: Top Statutory PIT Rate, 1981-2005

47

2030

4050

6070

%

1980 1985 1990 1995 2000 2005

High income Upper middleLower middle Low income

Unweighted average

2030

4050

6070

%1980 1985 1990 1995 2000 2005

High income Upper middleLower middle Low income

GDP-weightedFigure 2: Top Statutory PIT Rate by Income Group, 1981-2005

48

010

2030

4050

60%

1980 1985 1990 1995 2000 2005

Top statutory PIT MTR at 4y MTR at 3yMTR at 2y MTR at y

Marginal rate

010

2030

4050

60%

1980 1985 1990 1995 2000 2005

ATR at 4y ATR at 3yATR at 2y ATR at y

Average rateFigure 3: Actual PIT Rates, 1981-2005

Notes: Both marginal and average rates are weighted by GDP in constant 1990 U.S. dollars; y denotes the level of gross income equivalent to a country’s GDP per capita.

49

0

48

1216

2024

1980 1985 1990 1995 2000 2005

High income Upper middleLower middle Low income

Unweighted

04

812

1620

24

1980 1985 1990 1995 2000 2005

High income Upper middleLower middle Low income

WeightedFigure 4: Number of Brackets by Income Group, 1981-2005

50

4

68

1012

1416

%

1980 1985 1990 1995 2000 2005

All Income Taxes/GDPPIT/GDP

A. General Government, OECD Countries

02

46

810

%

1980 1985 1990 1995 2000 2005

High income Upper middleLower middle Low income

B. PIT, Central Government, All CountriesFigure 5: PIT Revenue as % of GDP, 1981-2005

Notes: Both panels report unweighted means by year. Panel A is based on the OECD Data Source for the general government that includes central as well as local and state governments. All income taxes are taxes on income, profits, and capital gains. The number of OECD countries ranges from 24 in 1981 to 29 in 2005 for all income taxes and 25 in 1981 to 29 in 2005 for PIT. Panel B is based on the OECD Data Source and IMF Government Finance Statistics for the consolidated central government only. The number of countries in Panel B ranges from 59 to 80 per year.

51

Appendix Table A1: Summary Statistics of Main PIT Variables by Year

Unweighted Weighted Top PIT

Rate Top PIT

Rate Number of Brackets MRP1 ARP1

1981 46.775 62.009 14.674 0.110 0.084 (23.775) (14.006) (6.570) (0.042) (0.032) 1982 48.192 57.265 13.245 0.096 0.080 (23.364) (13.796) (6.340) (0.034) (0.029) 1983 47.913 57.918 12.984 0.098 0.079 (23.219) (13.440) (6.477) (0.034) (0.028) 1984 45.930 56.736 12.503 0.098 0.078 (23.469) (12.289) (6.256) (0.034) (0.027) 1985 45.102 56.503 12.413 0.099 0.080 (22.959) (12.234) (6.199) (0.035) (0.027) 1986 44.083 56.280 12.246 0.097 0.078 (22.511) (12.183) (5.994) (0.033) (0.027) 1987 41.307 50.635 8.732 0.088 0.073 (21.851) (13.054) (5.631) (0.031) (0.026) 1988 40.323 47.153 6.760 0.079 0.066 (21.135) (15.881) (3.806) (0.031) (0.024) 1989 39.238 44.139 6.104 0.077 0.065 (20.771) (15.195) (3.740) (0.033) (0.024) 1990 37.372 42.345 5.818 0.078 0.065 (19.527) (13.471) (3.646) (0.034) (0.023) 1991 36.139 42.937 5.534 0.078 0.062 (19.105) (12.807) (3.712) (0.035) (0.022) 1992 34.977 41.566 5.260 0.078 0.061 (18.234) (12.714) (3.747) (0.033)  (0.024) 1993 34.047 40.689 5.108 0.076 0.060 (17.306) (12.223) (3.528) (0.034) (0.025)

52

Table A1 Cont’d: Summary Statistics of Main PIT Variables by Year Unweighted Weighted Top PIT

Rate Top PIT

RateNumber of Brackets MRP1 ARP1

1994 33.401 42.830 5.333 0.076 0.060 (16.709) (10.800) (2.758) (0.030) (0.024) 1995 33.024 42.786 5.554 0.067 0.052 (16.198) (10.478) (2.765) (0.024) (0.023) 1996 32.813 42.894 5.362 0.072 0.058 (16.257) (10.331) (2.715) (0.026) (0.023) 1997 32.785 42.463 5.195 0.074 0.058 (15.954) (10.370) (2.091) (0.027) (0.024) 1998 32.657 41.972 5.052 0.073 0.058 (15.875) (10.367) (1.910) (0.028) (0.024) 1999 32.335 40.271 4.907 0.072 0.057 (15.627) (9.589) (1.876) (0.027) (0.024) 2000 31.910 39.936 4.851 0.070 0.058 (15.463) (9.331) (1.855) (0.023) (0.024) 2001 31.224 39.346 5.049 0.069 0.055 (15.159) (9.328) (1.883) (0.023) (0.022) 2002 30.593 38.727 5.103 0.070 0.058 (14.772) (9.225) (1.864) (0.024) (0.024) 2003 29.914 37.353 5.073 0.070 0.055 (14.720) (9.106) (1.853) (0.025) (0.023) 2004 29.825 37.062 5.050 0.072 0.054 (14.611) (9.085) (1.862) (0.028) (0.024) 2005 29.011 36.434 5.031 0.068 0.052 (14.492) (8.815) (1.869) (0.024) (0.022) N 3613 3613 3143 3082 3082

Notes: Standard deviations are in parentheses. Weight is GDP in constant 1990 U.S. dollars. MRP1 and ARP1 are marginal and average tax rate progressions up to an income level equivalent to four times GDP per capita. All variables are described in the text.